Before I signed, Tesla gave me a production estimate for my roof: 26,992 kWh a year. That number drives almost every payback calculation a salesperson shows you, so I wanted to know how close it came. I now have almost three years of monthly numbers from the Tesla app, a weather model to compare them against, and a few service visits to explain the bad months.
The short answer: over a full year my system beat Tesla’s estimate by about 15%, but the estimate got the shape of the year badly wrong. Winter was more than double what Tesla predicted, and summer came in lower.
The system
A home in St. Augustine with 58 Tesla 425 W panels (24.65 kW DC) on six roof faces, all at a 27° pitch. Nineteen panels face east (85°), twenty-three face west (265°), twelve face almost due north (355°) and only four (MP7) face south. The roof is mostly east/west/north — I wrote about that in a mostly east/west/north roof, with only four panels facing south. Three solar inverters do the conversion: one inside each of my two Powerwall+ units and one standalone 7.6 kW Tesla inverter. It was installed in November 2023 and started producing in December 2023.
Year by year
| Year | Solar produced | Measured ÷ weather model |
|---|---|---|
| 2024 | 29,100 kWh | 0.94 |
| 2025 | 31,300 kWh | 0.99 |
| 2026 to Oct 23 | 26,400 kWh | 0.96 |
2024 includes about two months when one inverter was down (see below). 2026 is a partial year.
Lifetime production through October 23, 2026 is 87,327 kWh on my Home Assistant lifetime counter, and the monthly Tesla app totals add up to 87,410 kWh — close enough that I trust both.
Measured vs. Tesla’s estimate
The cleanest comparison is a full 12 months with no hardware problems. November 2024 through October 2025 fits that.
Two years vs. the estimate, season by season
Tesla’s estimate vs. what my system measured, kWh per season
Measured: Tesla app monthly totals (2 significant figures). Estimate: Tesla’s pre-install estimate for this system. Year totals: 31,100 kWh (Nov 2024–Oct 2025, +15%) and 30,200 kWh (Nov 2025–Oct 23, 2026, partial, +12%) vs. Tesla’s 26,992 kWh. Nov 2023–Oct 2024 is left out: December 2023 was a partial first month and one of my three inverters was effectively offline in January–February 2024.
| Season | Measured | Tesla estimate | Difference |
|---|---|---|---|
| Nov–Jan | 5,400 kWh | 2,268 kWh | 2.4× |
| Feb–Apr | 8,100 kWh | 6,582 kWh | +23% |
| May–Jul | 9,900 kWh | 11,339 kWh | −13% |
| Aug–Oct | 7,700 kWh | 6,803 kWh | +13% |
| Year | 31,100 kWh | 26,992 kWh | +15% |
The pattern repeated the next year: November 2025 through January 2026 made 5,500 kWh, and May through July 2026 made 9,900 kWh again. So this isn’t one lucky winter.
My best guess at why: Florida winters are dry and clear, and panels make more power when they’re cool. Summers bring afternoon thunderstorms and roof temperatures that cost real output. My weather model puts the heat loss at about 2,900 kWh a year, roughly 4% in December and January but over 10% from May to August. An estimate that assumes summer is simply “more sun” will miss both effects.
Why this matters: if you’re on net metering, the annual total is what pays you back, so beating the estimate by 15% is good news. But if you size a system to cover summer air-conditioning month by month, Tesla’s summer numbers will look better than reality.
Month by month against the weather
An annual estimate can’t tell you whether a given month was good or bad, because weather swings a lot. So I also run a weather model: it takes the actual hourly sun and temperature for St. Augustine and calculates what my exact roof layout should have produced. Dividing measured by modeled gives a ratio that shows when something is wrong.
Measured vs. what the weather says I should have made
kWh per month
Measured: Tesla app. Model: pvlib with Open-Meteo ERA5 weather for St. Augustine, my panel layout, the panel temperature coefficient and inverter limits. October 2026 runs through Oct 23.
Show the numbers
| Month | Measured kWh | Modeled kWh | Measured ÷ model |
|---|---|---|---|
| 2023-12 | 510 | 1,474 | 0.35 |
| 2024-01 | 904 | 1,642 | 0.55 |
| 2024-02 | 1,242 | 2,023 | 0.61 |
| 2024-03 | 2,671 | 2,704 | 0.99 |
| 2024-04 | 3,412 | 3,354 | 1.02 |
| 2024-05 | 3,592 | 3,516 | 1.02 |
| 2024-06 | 3,322 | 3,436 | 0.97 |
| 2024-07 | 3,222 | 3,365 | 0.96 |
| 2024-08 | 2,914 | 3,006 | 0.97 |
| 2024-09 | 2,154 | 2,298 | 0.94 |
| 2024-10 | 2,194 | 2,210 | 0.99 |
| 2024-11 | 1,873 | 1,855 | 1.01 |
| 2024-12 | 1,578 | 1,590 | 0.99 |
| 2025-01 | 1,880 | 1,770 | 1.06 |
| 2025-02 | 1,950 | 1,828 | 1.07 |
| 2025-03 | 2,809 | 2,889 | 0.97 |
| 2025-04 | 3,417 | 3,503 | 0.97 |
| 2025-05 | 3,356 | 3,426 | 0.98 |
| 2025-06 | 3,208 | 3,340 | 0.96 |
| 2025-07 | 3,295 | 3,389 | 0.97 |
| 2025-08 | 2,903 | 3,072 | 0.94 |
| 2025-09 | 2,594 | 2,646 | 0.98 |
| 2025-10 | 2,192 | 2,306 | 0.95 |
| 2025-11 | 2,112 | 1,971 | 1.07 |
| 2025-12 | 1,605 | 1,550 | 1.04 |
| 2026-01 | 1,824 | 1,789 | 1.02 |
| 2026-02 | 2,085 | 2,099 | 0.99 |
| 2026-03 | 2,697 | 2,792 | 0.97 |
| 2026-04 | 3,152 | 3,207 | 0.98 |
| 2026-05 | 3,338 | 3,410 | 0.98 |
| 2026-06 | 3,276 | 3,512 | 0.93 |
| 2026-07 | 3,288 | 3,620 | 0.91 |
| 2026-08 | 2,828 | 3,014 | 0.94 |
| 2026-09 | 2,458 | 2,610 | 0.94 |
| 2026-10 | 1,472 | 1,516 | 0.97 |
The model runs a little low in winter and a little high in summer, so the honest way to read it is to compare the same month across years. Most months land within a few percent of each other year to year — which is about the rounding precision of the Tesla app (it reports two significant figures, so 3,300 kWh really means 3,250–3,349).
The peak I’ve recorded is 19.57 kW, on a clear October afternoon. In that same week a cloudy day made 39.8 kWh and a clear day made 93.5 kWh — so “how much does it make a day” depends heavily on weather.
Service and reliability
Here is everything that went wrong, and what it cost in production. Service was handled by Tesla’s Orlando-area team.
January–February 2024: inverter fan
My first two full months were weak. Tesla’s daily history shows measured ÷ model of 0.585 for January–February 2024 (0.55 in January, 0.61 in February), when a healthy winter here runs about 1.04. The standalone inverter handles about 44% of the winter output, and if it produced nothing while everything else was normal, the model predicts 0.584 — a match to the third decimal. Tesla replaced that inverter’s fan in February 2024, and the daily numbers were back to normal by about March 6. Measured against a healthy 1.04, January–February lost about 1,700 kWh. December 2023, my startup month, ran even lower (0.34), and I don’t have a service record that explains it, so I haven’t counted it here.
2026: the standalone inverter again
The same standalone inverter later lost cooling and was reported as underproducing, and Tesla replaced the whole unit on June 25, 2026. I pulled Tesla’s daily and hourly history to measure what the swap did. In the 30 days before it, the system ran at 0.946 of the weather model; in the 30 days after, 0.903. That looks worse, but the same calendar windows drop every year when the summer storms arrive: 0.981 to 0.931 in 2025, and 1.011 to 0.934 in 2024. Measured against that normal seasonal drop, the swap changed output by about +1% (compared with 2025) to +3% (compared with 2024), which is within the noise. March to May 2026 matched March to May 2025 exactly (0.976 both years).
The hourly data doesn’t show clipping before the swap either. Between April 1 and June 24, 2026, the system topped 18 kWh in an hour 16 times, more than in the same stretch of 2025, and on sunny spring days afternoon output tracked the model as well as mornings did (the same as 2024 and 2025). The measured result: no detectable loss before the swap and no detectable gain after it.
Summer 2026 was still weak. June to September ran at 0.929 vs. 0.964 in 2025, about 3.6% or 450 kWh lower, and July 2026 (0.91) was my worst month relative to the weather since early 2024. The shortfall is mostly in the afternoons, and it started in early June, before the swap, so the inverter doesn’t explain it. Stormier afternoons than my weather data captures are the most likely suspect, but I can’t prove it. October 1–23, 2026 is running at 0.97 vs. 0.92 a year earlier.
Degradation
Panels lose a little output each year. My numbers suggest something around 1% a year, but with a margin of error of 2–3%, so I can’t honestly separate normal aging from weather and rounding yet. Ask me in a few more years.
Practical tips
- Ask for the monthly estimate, not just the annual one. The annual total was conservative for me; the monthly shape was wrong.
- Download your monthly data every few months. The Tesla app lets you export it. Two bad months in a row is the signal to call service.
- Compare the same month year to year. Comparing March to July tells you nothing; comparing March 2025 to March 2026 does.
- Watch the inverter that carries the most panels. When my standalone inverter stopped, I lost close to half my winter output.
- Don’t panic over 2–3%. That’s inside the app’s rounding.
Next: what all that production is worth in dollars — see what my system really cost and how fast it’s paying for itself.